Fine-tuning with Contextual Information Beyond Prompt-Response Pairs: Possible?

Just turn data into a PDF and attach to prompt the query as you wish and by the way your chats have memory and it does learn as you feed it info that’s the training